Bug: Test Connection in Embedding Search doesn't work when entering all required info. Add dropdown presets for model selection.

[apoapostolov]

Bug Report

The "Test Connection" button in Embedding Search Settings doesn't properly test the connection when all required fields are entered.

Current Behavior

When I fill in:

• API URL:
• API Key: Valid OpenAI key (sk-...)
• Model:

And click "Test Connection", it doesn't give any meaningful feedback about whether the connection actually works or not.

The UI just shows a spinner then returns without indicating:

• Whether the connection succeeded
• What error occurred (if any)
• The model used
• Latency/response time

Expected Behavior

The "Test Connection" feature should:

1. Actually call the embedding API with the provided credentials

2. Return clear feedback:

   • ✅ "Connection successful! Latency: 245ms"
   • ❌ "Connection failed: Invalid API key"
   • ⚠️ "Model not found: text-embedding-3-small"

3. Show the test details:

   • Model used
   • Response time
   • Embedding dimensions (if available)

Feature Request: Dropdown Presets

Additionally, when selecting the Embedding Model, instead of typing the model name manually (which is error-prone), provide a dropdown with common presets:

OpenAI Presets:

• (default)

Custom option: Allow manual typing for advanced users

This would make it much harder to:

• Typos (e.g., text-embeding vs text-embedding)
• Forgetting the exact model names
• Confusing which models are available

Why this matters

Without proper connection testing and model presets, users can't confidently configure embedding search, which means the powerful semantic search feature is effectively unusable.

Environment

• Memory Viewer version: 1.2.0
• Browser: Chrome/Brave/Firefox
• OS: Tested on Linux

Steps to Reproduce

1. Go to Settings → Embedding Search
2. Fill in API URL, API Key, Model
3. Click "Test Connection"
4. Observe that no clear success/failure feedback is shown